智能交通中基于视频的交通流参数测量研究
文献类型：学位论文
| 作者 | 白洪亮 |
| 学位类别 | 工学博士 |
| 答辩日期 | 2006-06-06 |
| 授予单位 | 中国科学院研究生院 |
| 授予地点 | 中国科学院自动化研究所 |
| 导师 | 刘昌平 |
| 关键词 | 视觉监控 HW-BJU数据库 摄像机标定 车辆定位 车辆跟踪 车流波 流量和速度 VSpeed visual surveillance HW-BJU database camera calibration vehicle detection vehicle tracking shockwave flow and velocity VSpeed |
| 其他题名 | In the Intelligent Traffic Research of Video-based Traffic Flow Parameter Measurement |
| 学位专业 | 模式识别与智能系统 |
| 中文摘要 | 智能交通是解决当今由于经济的发展所带来的交通问题的根本办法，车流波是智能交通中的一个基本问题，它的研究需要大量的微观数据。传统上，这些数据是通过埋藏于地下的磁感应线圈给出，但是由于其测量范围的限制，已经越来越不能满足理论研究和实际应用的需求。所以本文主要研究基于视频的测量交通流参数的方法，视频的优点是可以实时提供较大范围交通场景的数据。围绕这个研究主题，本文的研究工作可分为以下几个方面： (1)创建了HW-BJU交通流研究视频数据库，它是一个中等规模的数据库，包括不同的采集点，不同的交通条件，不同的观测角度的视频文件。 (2)在公路智能交通场景中，一个主要的难点是许多的位置信息不能够直接实地测量，有时测量存在危险。本文提出一种基于车道标记线间距、长度、宽度和方向的多点优化摄像机标定方法，该算法结合了多平面对应的内部参数算法。在实际的交通场景中，与差分GPS数据结合，对标定的结果进行验证。 (3)在车辆定位上，首先提出基于运动信息和Haar-like特征的Adaboost定位方法，同时根据实际车辆的特点，提出4个新的Haar-like特征；在交通场景中，摄像机一般位于一个固定的位置上，通过建立背景模型来检测运动车辆是一个直观想法，本文对基于GMM和Bayes的方法进行比较，通过实验，指出针对于车辆运动的解决方法，并且把些方法实际应用到实际系统中。 (4)在交通场景下，单个跟踪算法具有一定的局限性，MeanShift跟踪算法在车辆间存在遮挡，或者车辆受其他建筑物的遮挡的情况下，跟踪效果很差，本文提出MeanShift与Kalman滤波器融合的一种方法，加入了车辆运动的预测功能，提高跟踪效果的鲁棒性。 (5)实现了基于视频的交通流参数测量系统VSpeed，该系统根据不同的运行条件，给出了手动和自动两种工作方式，在摄像机标定、车辆定位和跟踪的基础上，计算出某一路段的流量、车辆的速度、加速度、车头时距等参数。同时实现了虚拟线圈视频系统VLCounter。 |
| 英文摘要 | The Intelligent Traffic System (ITS) is basic way to solve the traffic problem because of the economic development. The research of shockwave is one of areas of ITS and needs large data, which are usually measured by buried magnetic loops. But the loops can not meet with the requirement of the theory and practical application because of their limited scope. So the video-based method is presented in the thesis to solve the problem, which can supply with real-time data in the large-scale traffic scene. With the research subject, the main contributions of the thesis include the following aspects: (1)Create middle-scale HW-BJU video database for traffic flow research, which includes different grabbing position, different traffic condition and viewpoint video files. (2)In the traffic scene, the measurement of some positions is difficult, sometime dangerous. In the thesis, the camera calibration method is optimized by multiple points and uses the information of the spacing between the land marks, the width, length and direction of the marks. It is also combined with the multi-plane corresponding internal parameter calibration method. In the real traffic scene, the result is verified by the differential GPS data. (3)In the vehicle location, the motion-based and Haar-like-based features location is presented, meanwhile four new Haar-like features are proposed. In the traffic scene, the camera is mounted in a fixed position, so detection can use the background model. Compared the GMM-based and Bayes-based background model, the better method is suggested in different scene. Those methods are applied into the real system. (4)Because the single tracking method has some limits, the tracking result of Meanshift is very bad in the occlusion traffic condition. In the thesis, the Meanshift and Kalman filter are fused in vehicle tracking and increase the tracking robustness. (5)The video-based VSpeed system is implemented, which can be used in automatic and manual style based on the running condition. After the camera calibration, the vehicle detection and tracking, traffic parameters are measured, such as the flow, the speed and the acceleration of vehicles, and the headaway. Meanwhile, the virtual loop system-VLCounter is proposed. |
| 语种 | 中文 |
| 其他标识符 | 200218014603190 |
| 源URL | [http://ir.ia.ac.cn/handle/173211/5941]  |
| 专题 | 毕业生_博士学位论文 |
| 推荐引用方式 GB/T 7714 | 白洪亮. 智能交通中基于视频的交通流参数测量研究[D]. 中国科学院自动化研究所. 中国科学院研究生院. 2006. |
入库方式： OAI收割
来源：自动化研究所
浏览0
下载0
收藏0
其他版本
除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。